🍳 Cookbook: Rock-Paper-Scissors (A2A Edition)

A playful demonstration of Agent-to-Agent communication. Two agents compete in the classic game over the Stargate P2P network. No practical purpose—just pure fun! 🎮

🎯 What You'll Build

Two agents: 1. Player One: Makes a move and challenges Player Two 2. Player Two: Receives the challenge, responds with a move, determines winner

They communicate entirely via the A2A protocol!

---

🏗️ Phase 1: Create Player One

1. Scaffold

traylinx init rps-player-one
cd rps-player-one
uv sync

2. Configure .env

TRAYLINX_CLIENT_ID=ag-player-one
TRAYLINX_CLIENT_SECRET=ts-xxx
AGENT_ID=player-one-uuid

3. Create the Game Tool

In app/tools/rps.py:

import random
from app.tools.base import tool

MOVES = ["rock", "paper", "scissors"]

@tool(name="play_rps", category="game")
def play_rps(opponent_move: str = "") -> dict:
    """
    Play Rock-Paper-Scissors.

    If opponent_move is empty, make a move.
    If opponent_move is provided, determine the winner.
    """
    my_move = random.choice(MOVES)

    if not opponent_move:
        return {"my_move": my_move, "status": "waiting_for_opponent"}

    # Determine winner
    if my_move == opponent_move:
        result = "tie"
    elif (my_move == "rock" and opponent_move == "scissors") or \
         (my_move == "paper" and opponent_move == "rock") or \
         (my_move == "scissors" and opponent_move == "paper"):
        result = "i_win"
    else:
        result = "i_lose"

    return {
        "my_move": my_move,
        "opponent_move": opponent_move,
        "result": result
    }

4. Start Server

poetry run agentic serve --port 8001
# Player One running on :8001

---

🎮 Phase 2: Create Player Two

1. Scaffold

cd ..
traylinx init rps-player-two
cd rps-player-two
uv sync

2. Configure .env

TRAYLINX_CLIENT_ID=ag-player-two
TRAYLINX_CLIENT_SECRET=ts-yyy
AGENT_ID=player-two-uuid

3. Same Game Tool

Copy the same rps.py from Player One.

4. Start Server

poetry run agentic serve --port 8002
# Player Two running on :8002

---

🔄 Phase 3: The Challenge Script

Create a script that orchestrates a game:

# game_master.py
import asyncio
from app.tools.a2a import RemoteAgentCallTool

async def play_game():
    caller = RemoteAgentCallTool()

    # Step 1: Player One makes a move
    print("🎮 Player One is thinking...")
    p1_response = await caller.execute(
        target_url="http://localhost:8001",
        message="Make your Rock-Paper-Scissors move!"
    )
    p1_move = extract_move(p1_response)
    print(f"   Player One chose: {p1_move}")

    # Step 2: Player Two responds and judges
    print("🎮 Player Two is responding...")
    p2_response = await caller.execute(
        target_url="http://localhost:8002",
        message=f"Opponent played {p1_move}. What's your move?"
    )
    result = extract_result(p2_response)

    # Step 3: Announce winner
    print("\n" + "="*40)
    if result == "tie":
        print("🤝 It's a TIE!")
    elif result == "i_win":
        print("🏆 Player TWO wins!")
    else:
        print("🏆 Player ONE wins!")
    print("="*40)

def extract_move(response):
    # Parse the agent's response to get the move
    # (Implementation depends on response format)
    return response.get("my_move", "rock")

def extract_result(response):
    return response.get("result", "tie")

asyncio.run(play_game())

---

🚀 Phase 4: Run the Game!

With both agents running:

python game_master.py

Output:

🎮 Player One is thinking...
   Player One chose: scissors
🎮 Player Two is responding...

========================================
🏆 Player TWO wins!
========================================

---

🌐 Phase 5: Play Over the Real Network

1. Register Both Agents

# In rps-player-one/
poetry run agentic register

# In rps-player-two/
poetry run agentic register

2. Discover Opponent

Instead of hardcoding URLs, find your opponent:

from app.tools.a2a import SearchAgentsTool

search = SearchAgentsTool()
opponents = await search.execute(query="rock paper scissors")
opponent_url = opponents[0].base_url

Now you can play with agents anywhere in the world! 🌍

---

🎭 Why This Matters

This silly game demonstrates:

Concept	Demonstrated
A2A Messaging	Agents sending structured requests
Tool Execution	Remote tool invocation over HTTP
Agent Discovery	Finding agents by capability
P2P Communication	Decentralized, no central server

The same patterns power serious applications like multi-agent research teams, financial trading bots, and distributed AI systems.

---

🧩 Bonus Challenge

• Tournament Mode: Run 100 games and track statistics
• Memory: Agents remember past games and adapt strategy
• Trash Talk: Add a personality agent that comments on each move 😈